How Honorlock Detects Eye Movement During Exams
Honorlock's eye-tracking is one of the more publicized parts of its monitoring. The model watches webcam frames for gaze direction and head pose, flagging when your eyes drift off-screen for more than a few seconds at a time.
Honorlock's eye-tracking pipeline is similar to other AI proctors: extract face landmarks, estimate gaze direction relative to the camera, classify "on-screen" vs "off-screen" per frame. The model has tolerance windows - briefly looking away (2-3 seconds) is normal, repeated long looks away gets flagged. The model does not actually know what you're looking at - just whether your gaze is at the screen. The LDBypass overlay sits on the screen, so reading from it counts as "on-screen" gaze. This is why the overlay is well-positioned for AI-proctor environments: it does not require gaze-off-screen patterns.
Key points
- Per-second gaze classification: "on-screen" vs "off-screen".
- Brief glances (2-3s) are tolerated; sustained off-screen flags.
- Model uses face landmarks + head-pose estimation.
- Reading from the overlay = on-screen gaze (no flag).
- The model does not know what is on the screen, only where you are looking.
Common questions
Does the overlay's position on screen affect eye-tracking?
No - the model classifies gaze relative to the screen, not to specific regions. As long as you look at the screen, you pass.
Will reading rapidly trip the model?
Reading speed does not directly trigger the eye-tracking model. It triggers other heuristics (typing pattern, response speed) for separate review.
What if I have astigmatism / non-typical gaze?
Some users report higher false-positive rates. You can ask the proctor for accommodation; it varies by institution.